Back to the original app.py

app.py

@@ -6,10 +6,15 @@ load_dotenv()
 
 import os
 import sys
+import getpass
+import nest_asyncio
+# import pandas as pd
 import faiss
 import openai
 
 import chainlit as cl  # importing chainlit for our app
+# https://docs.chainlit.io/api-reference/step-class#update-a-step
+# DEPRICATED: from chainlit.prompt import Prompt, PromptMessage  # importing prompt tools
 
 import llama_index
 from llama_index.core import Settings
@@ -23,6 +28,10 @@ from llama_index.embeddings.openai import OpenAIEmbedding
 from llama_index.postprocessor.flag_embedding_reranker import FlagEmbeddingReranker
 from llama_parse import LlamaParse
 
+from openai import AsyncOpenAI  # importing openai for API usage
+
+os.environ["CUDA_VISIBLE_DEVICES"] = ""
+# GET KEYS
 LLAMA_CLOUD_API_KEY= os.getenv('LLAMA_CLOUD_API_KEY')
 OPENAI_API_KEY=os.getenv("OPENAI_API_KEY")
 
@@ -32,6 +41,8 @@ os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")
 # os.environ["WANDB_API_KEY"] = getpass.getpass("WandB API Key: ")
 """
 
+nest_asyncio.apply()
+
 # PARSING the pdf file
 parser = LlamaParse(
     result_type="markdown",
@@ -42,7 +53,7 @@ parser = LlamaParse(
 
 nvidia_docs = parser.load_data(["./nvidia_2tables.pdf"])
 # Note: nvidia_docs contains only one file (it could contain more). nvidia_docs[0] is the pdf we loaded.
-# print(nvidia_docs[0].text[:1000])
+print(nvidia_docs[0].text[:1000])
 
 # Getting Settings out of llama_index.core which is a major part of their v0.10 update!
 Settings.llm = OpenAI(model="gpt-3.5-turbo")
@@ -54,11 +65,11 @@ node_parser = MarkdownElementNodeParser(llm=OpenAI(model="gpt-3.5-turbo"), num_w
 
 nodes = node_parser.get_nodes_from_documents(documents=[nvidia_docs[0]])
 # Let's see what's in the metadata of the nodes:
-# for nd in nodes:
-#   print(nd.metadata)
-#   for k,v in nd:
-#     if k=='table_df':
-#       print(nd)
+for nd in nodes:
+  print(nd.metadata)
+  for k,v in nd:
+    if k=='table_df':
+      print(nd)
 # Now we extract our `base_nodes` and `objects` to create the `VectorStoreIndex`.
 base_nodes, objects = node_parser.get_nodes_and_objects(nodes)
 
@@ -79,11 +90,24 @@ recursive_index_faiss = VectorStoreIndex(nodes=base_nodes+objects, storage_conte
 # We'll need to do a couple steps:
 # 1. Initalize our reranker using `FlagEmbeddingReranker` powered by the `BAAI/bge-reranker-large`.
 # 2. Set up our recursive query engine!
+
 reranker = FlagEmbeddingReranker(
-    top_n=1,
+    top_n=5,
     model="BAAI/bge-reranker-large",
 )
 
+recursive_query_engine = recursive_index_faiss.as_query_engine(
+    similarity_top_k=15,
+    node_postprocessors=[reranker],
+    verbose=True
+)
+
+"""
+# Create pandas dataframe to store query+generated response+added truth
+columns=["Query", "Response", "Truth"]
+gen_df = pd.DataFrame(columns=columns,dtype='str')
+"""
+
 # ChatOpenAI Templates
 system_template = """Use the following pieces of context to answer the user's question.
 If you don't know the answer, say that you don't know, do not try to make up an answer. 
@@ -92,44 +116,29 @@ The "SOURCES" part should be a reference to the source inside the document from
 You are a helpful assistant who always speaks in a pleasant tone! """
 
 user_template = """ Think through your response step by step."""
-    
+
 #user_query = "Who are the E-VP, Operations - and how old are they?"
-def resursive_fn(reranker):
-    recursive_query_engine = recursive_index_faiss.as_query_engine(
-        similarity_top_k=1,
-        node_postprocessors=[reranker],
-        verbose=True
-    )
 
-    return recursive_query_engine
+#response = recursive_query_engine.query(system_template + user_query + user_template)
+
+#str_resp ="{}".format(response)
 
-recursive_fn_val = resursive_fn(reranker)
 
-@cl.on_chat_start 
-async def start_chat():
-    print("A new chat session has started!") 
-    cl.user_session.set("recursive_query_engine", recursive_fn_val)
-    
+def retriever_resp(prompt):
+    import time
+    response = "this is my response"
+    time.sleep(5)
+    return response
 
 @cl.on_message  # marks a function that should be run each time the chatbot receives a message from a user
 async def main(message: cl.Message):
     settings = cl.user_session.get("settings")
-    
-    user_query = message.content
-    print("inside on_message - user_query: ",user_query)
-    prompt=system_template + user_query + user_template
-
-    recursive_query_engine = cl.user_session.get("recursive_query_engine")
-    print("inside on_message - recursive_query_engine: ",recursive_query_engine)
-
-    response = await recursive_query_engine.query(prompt)
-
-    print("inside on_message - response: ",response)
 
+    user_query = message.content
+#    prompt = system_template+user_query+user_template
+    response = recursive_query_engine.query(system_template + user_query + user_template)
+#    response = retriever_resp(prompt)
+#    print("AAA",user_query)
     str_resp ="{}".format(response)
-    
-    # response = await recursive_fn_call(recursive_query_engine, system_template, user_template, user_query=user_query)
     msg = cl.Message(content= str_resp)
-    print("inside on_message - after msg: ",msg)
-
     await msg.send()